Multivariate second order Poincaré inequalities for Poisson functionals (Q2279325)

Given a Poisson process \(\eta\) over a measurable space with a \(\sigma\)-finite intensity measure, a random variable \(F\) is called a Poisson functional if there is a measurable map \(f\) such that \(F=f(\eta)\). For a vector \((F_1,\ldots,F_m)\) of such Poisson functionals, the main results of this paper are explicit error bounds in the approximation of their distribution by an \(m\)-dimensional Gaussian distribution. Such inequalities are referred to as second order Poincaré inequalities. The authors give error bounds in distances based on smooth test functions, but also in the distance \[ d_{\mathrm{convex}}(Y,Z)=\sup_{h\in\mathcal{I}_m}|\mathbb{E}h(Y)-\mathbb{E}h(Z)|\,, \] where \(\mathcal{I}_m\) is the set of all indicator functions of measurable convex sets in \(\mathbb{R}^m\). The proofs are based on Stein's method for multivariate Gaussian approximation and Malliavin calculus on the Poisson space, with a suitable smoothing argument needed to establish bounds in the distance \(d_{\mathrm{convex}}\). These results are multivariate analogues of the corresponding univariate inequalities of [\textit{G. Last} et al., Probab. Theory Relat. Fields 165, No. 3--4, 667--723 (2016; Zbl 1347.60012)]; as in that earlier work, they give error bounds expressed in terms of integrated moments of first and second order difference operators. These inequalities give rates of convergence which are apparently optimal. Within their proof, the authors give new bounds on the second derivatives of the solution to the Stein equation for multivariate Gaussian approximation, which may well prove to be useful in other applications of the technique. The main results of the paper are illustrated with several applications, including (multivariate) central limit theorems for first order Poisson integrals and intrinsic volumes of Boolean models.